[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: OT: performance problems.

> Linux piper 2.4.9 #1 Tue Sep 11 15:39:28 CEST 2001 i686 unknown
> 16:45:25 up 13 days,  1:17,  7 users,  load average: 3.40, 3.56, 3.70
> 84 processes: 83 sleeping, 1 running, 0 zombie, 0 stopped     
> CPU states:   0.8% user,  22.3% system,   1.0% nice,  75.9% idle

> Mem:    505844K total,   502788K used,     3056K free,    11456K buffers
> Swap:   996020K total,    31172K used,   964848K free,   438552K cached
> however, i am continuously having troubles. for instance, in a typical
> situation, i'd have windowmaker running with four terms, xmms playing
> some 192kbps MP3s, some ssh sessions into it, and an rsync, bzip/gzip,
> or make-kpkg process running. i am not usually interactively using X.
> in such a situation, xmms (or mpg123 without X, it doesn't matter)
> continuously skips on MP3s and it's *very* annoying. i even went as far
> as to renice xmms to -20 *and* rsync/bzip/gzip/make-kpkg to 20, but it
> doesn't really help.

First of all (stupid question): you aren't using esd or something alike?
Because although XMMS may get all the time it needs, esd may be starved...

Second: Even though one has -20 and the other 20 doesn't mean that the 
last one won't get anymore time... For example, I use distributed-net 
on some machines: Even though dnetc has nice 20 it ALWAYS has a minimum
of 5-10% CPU time. 
I know WinNT has somekind of deadlock protection, where it increases the
priority of a starved process every tick until it got scheduled again.

(on another PC):
cat /dev/zero | bzip2 > /dev/null

 1721 root      20 -20  7768 7768   372 R <  90.4  6.1   0:35 bzip2
 1702 daemon    19  19   460  460   384 R N   8.2  0.3   0:53 dnetc
 1720 root      -1 -20   372  372   308 S <   0.7  0.2   0:00 cat

CPU usage of dnetc NEVER goes under 8%...

Third: I have a similar computer (AMD 1.333 Ghz, 512MB 266Mhz DDR, 7200
IDE disk, kernel 2.4.x, MSI Turbo-Pro motherboard). I wanted to check 
if the CD-writer was working OK (plextor clone) so I did 
   cat /dev/hdc | md5sum 
on the original and copy. While I was doing that, load also got up to 3-4.
Looking at top, I saw that some kernel daemon (think kupdated) got *AlOT*
of CPU. The system also started responding slow (missing eth0 traffic,

Maybe it's normal, 
maybe it's a kernel bug (only seem to have it with AMD) / chipset bug?

 Just some thoughts...


Reply to: